[U-Boot-Users] Using DCACHE instead of internal SRAM as the initial stack on PPC440GP?

[Stefan Roese <sr@denx.de>]

Hi Shawn,

On Saturday 05 March 2005 03:07, Shawn Jin wrote:

> I'm experimenting to set up the initial stack in DCACHE instead of
> internal SRAM on Ebony. Our SoC only has a ppc440 core and no internal
> SRAM available.
>
> I made more changes than what I reported yesterday
> (http://sourceforge.net/mailarchive/message.php?msg_id=11062649). My
> understanding on the initial stack is:
> 1. CFG_INIT_RAM_ADDR is set to an address which nobody uses it. No
> backup RAM is required.
> 2. Config only this area cacheable. All others are caching inhibited.
>
> So I made the following changes on Ebony's uboot code.
> 1. set CFG_INIT_RAM_ADDR to 0x40000000, no memory or peripherals use
> this address space, and CFG_INIT_DCACHE_CS to 4. 
> 2. This address space is mapped to 0x1_4F00_0000, one of spared
> peripheral areas. My TLB is set as follows.
>
> (gdb) mon tlb 0 8
>  0 : 00 f0000000 256MB V0 -> 1_f0000000 U:0000 -I-G- XWRXWR
>  1 : 00 e0000000 256MB V0 -> 1_40000000 U:0000 -I-G- -WR-WR
>  2 : 00 40000000   4KB V0 -> 1_4f000000 U:0000 ----- XWRXWR

Tlb 1 and 2 seem to overlap in the physical addresses!

>  3 : 00 00000000 256MB V0 -> 0_00000000 U:0000 -I-G- XWRXWR
>  4 : 00 d0000000 256MB V0 -> 2_00000000 U:0000 -I-G- -WR-WR
>  5 : 00 80000000 256MB V0 -> 3_00000000 U:0000 -I-G- -WR-WR
>  6 : 00 00000000   1KB -0 -> 0_00000000 U:0000 ----- ------
>  7 : 00 00000000   1KB -0 -> 0_00000000 U:0000 ----- ------
>  8 : 00 00000000   1KB -0 -> 0_00000000 U:0000 ----- ------
>
> 3. Commented out internal SRAM initialization and TLB entries.
> 4. set up EBC bank 4 for this area.
> 	/*-----------------------------------------------------------------------
> */ /* Memory Bank x (nothingness) initialization  */
> 	/* used as temporary stack pointer for stage0  */
> 	/*-----------------------------------------------------------------------
> */ li	r4, xbcfg
> 	mtdcr	ebccfga, r4
> 	lis	r1, 0x0400
> 	ori	r1, r1, 0x0000
> 	mfdcr	r4, ebccfgd
> 	or	r4, r4, r1
> 	mtdcr	ebccfgd, r4		/* set ATC */
>
> 	li	r4, PBxAP
> 	mtdcr	ebccfga, r4
> 	lis	r4, 0x0380
> 	ori	r4, r4, 0x0480
> 	mtdcr	ebccfgd, r4
>
> 	addi	r4, 0, PBxCR
> 	mtdcr	ebccfga, r4
> 	lis	r4, 0x4F01		/* BAS=0x4F0, BS=000 (1MB) */
> 	ori	r4, r4, 0xE000	/* BU=11 (r/w), BW=11 (32 bits) */

You don't set BW to 11 but to 10. Please use 0x4f01f000 for the PBxCR value. I 
don't think this is the cause for the problem, but it can't hurt to correct 
this.

> 	mtdcr	ebccfgd, r4
>
> However initializing the stack still causes machine check exception. I
> really don't know what else I'm missing here. Any suggestions or hints
> are greatly appreciated!

Did you take a look at the error reporting registers (for example EBC0_BEAR / 
EBC0_BESR)?

Best regards,
Stefan